基于相对特征的音乐哼唱多句检索算法

摘要

Facing the huge music data, it is an important context for the study that how to build indexical architecture and improve the retrieve velocity in the music content-based retrieval system.This paper built the music feature and index on sentence, and the feature in the database was relative feature of music rhythm.But user would hum multi-sentence sometimes, the multi-sentence feature was needed.According to this problem, proposed and applied the solutions in the system.It included the conversion of the sing-sentence and multi-sentence feature and matching problem, and obtained the good retrieval result inthe system.The solution could be used in the retrieval system of the time series data with the similar structure too.%面对海量音乐数据,如何在基于内容检索时对其建立索引以提高检索速度是一个非常重要的研究内容.以句为单位对音乐内容提取特征并建立索引时,库中保存的是音乐旋律的相对特征,但用户在哼唱检索时,有时会哼唱一段包含多句的音乐来检索,这就需要生成多句特征.对此提出单句特征和多句转换及匹配问题的解决方案并应用于检索系统中,获得了较好的检索效果,相关成果也可应用于具有相似结构的时间序列数据的检索系统中.